Anthropic Expands Claude Text Watermark, but Two Older Models Have a Sampling Loophole

Anthropic is expanding its text watermark to four more Claude models: Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6, and Claude Sonnet 4.6. The company told API customers that newer models such as Claude Opus 5.5 already carry the mark, while Amazon Bedrock rollout may take a few extra days. The watermark is described as an imperceptible pattern in word choice that does not change meaning, quality, or readability.

How the loophole works

On Claude Opus 4.6 and Claude Sonnet 4.6, Anthropic says the watermark applies only when developers leave temperature, top_p, and top_k at their default values. These parameters control how much randomness the model uses when picking words, and a minor change to any one of them is enough to dodge the mark. Anthropic warns that customers who rely on the watermark for compliance should keep those settings at their defaults.

The email does not mention that caveat for Claude Mythos 5 or Claude Opus 4.7, which suggests the watermark applies to those models regardless of sampling settings. The loophole is still notable because Opus 4.6 remains Anthropic's best writing model, ahead of even Opus 5.5.

What the watermark does

Anthropic says the watermark adds no extra characters or tokens, does not affect pricing or request and response formats, and barely slows output speed. It also says the watermark contains no information about the user, their organization, or their conversations with Claude. Customers cannot turn it off, and it works on every platform that hosts these models, including Amazon Web Services, Google Cloud, and Microsoft Foundry.

Anthropic uses a variant of Google DeepMind's SynthID text method, which tweaks the randomness behind word selection to produce a statistically detectable pattern. A detection API is in private preview, and more details are available on Anthropic's content labeling information page.

EU deadline and industry context

Anthropic points to the EU AI Act's December deadline as the reason it is watermarking older models now. Every model released after Aug. 2, 2026 shipped with the watermark on day one, while older models are getting it in stages. Because the watermark is embedded at the model level, it works across hosting platforms.

OpenAI launched its own watermarking system, textGrain, in October 2026, but it only activates by default for ChatGPT and Codex users in the EU. API customers worldwide can turn it on or off, and it ships disabled. Google DeepMind open-sourced SynthID for text and built it into Gemini models.

Quality and legal questions

Whether watermarking truly leaves text quality untouched is disputed. Blogger John Gruber argues that no two synonyms mean exactly the same thing, so picking words by a watermark key instead of semantic fit inevitably degrades precision. Legal trade outlet Artificial Lawyer raises a separate concern for law firms, where clients who have banned AI use could discover the mark, and multiple watermarks layered inside one document could create complications.

Sources: The Decoder